Towards resilient pipeline infrastructure: lessons learned from failure analysis

Understanding the mechanisms of pipeline failures is crucial for identifying vulnerabilities in gas transmission pipelines and planning strategies to enhance the reliability and resilience of energy supply chains. Existing studies and the American Society of Mechanical Engineers' (ASME) Code for Pressure Piping primarily focus on corrosion, recommending inspections every 10 years to prevent incidents due to this time-dependent threat. However, these guidelines do not provide comprehensive regulation on the likelihood of incidents due to other causes, especially non-time-dependent events (i.e. do not provide any indication of the inspection frequency or the most likely time for an incident to occur). This study adopts an innovative approach adopting machine learning, particularly artificial neural networks (ANNs), to analyse historical pipeline failure data from 1970 to 2023. By analysing records from the US Pipeline & Hazardous Materials Safety Administration, the model captures the complexity of various degradation phenomena, predicting failure years and hazard frequencies beyond corrosion. This innovative approach allows adopting more informed preventive measures and response strategies, offering deep insights into incident causes, consequences, and patterns. The results provide practical insights for maintenance planning, offering an estimation of periods when a pipeline may be more susceptible to incidents based on various factors. However, since all models inherently present uncertainties, both in the data and the modelling process, these estimates should be interpreted as probabilistic assessments. This study provides operators with a strategic framework to prescriptively address potential vulnerabilities, thereby promoting sustained operational integrity and minimising the occurrence of unexpected events throughout the service life of pipelines. By expanding the scope of risk assessment beyond corrosion, this study significantly advances the field of pipeline safety and reliability, setting a new standard for comprehensive incident prevention. This study proposes a machine learning (ANNs) model to estimate the year of incidents in pipelines, considering multi-cause-and-effect relationships.12,182 pipeline incidents from 1970 to 2023 in the United States are analysed, considering both time-dependent and non-time-dependent hazards.The results offer probabilistic insights for a deeper understanding of pipeline failure dynamics, promoting risk mitigation.